A magazine where the digital world meets the real world.
On the web
- Browse by date
- Browse by topic
- Enter the maze
- Get our RSS feed
- Follow us on Twitter
- Resources for teachers
What is cs4fn?
- About us
- Contact us
- Privacy and cookies
- Copyright and contributions
- Links to other fun sites
- Complete our questionnaire and get a free magic download
Can computers help understand the brain?
Creating a brain inside a computer
For years a dream of computer scientists has been to build computers that can think. Many have looked at how the brain works and tried to model this in a computer: write computer programs that work like the brain does - so that maybe they would then start to think like a human! So far we haven't succeeded but it turns out that by trying we can help medics understand the way our brains work and maybe even help find cures. Rafal Bogacz, of the Department of Computer Science at the University of Bristol tells us more.
The human brain is made up about 100,000,000,000 information processing cells, called neurons. The neurons are connected by 'wires' that carry electrical signals, rather like the wires in a computer do. The total length of these 'wires' in a human brain is about 100,000 miles! That's half the distance between the earth and the moon.
Computer scientists have modelled systems like that in the brain using a mathematical construct called a Neural Network. A Neural Network is a bit like a set of pipes with water flowing through them. The pipes are connected by valves. Each valve has a set of pipes where the water enters (input pipes) and a set of pipes where the water leaves (output pipes). The amount of water that passes into the output pipes depends both on the valve and the amount of water on the input pipes. If you replace the water with electricity and the pipes with neurons you actually get something very similar to how the brain works!
Modelling the brain
Recently the focus has shifted. A new goal is help medics understand the brain using the computer science which people originally thought might create artificial intelligence. Many of the areas of the brain aren't completely understood. There are many parts of the brain that we don't understand why they are as they are. Understanding how individual parts of the brain work can be helpful in diagnosing diseases and creating cures.
One area of the brain which until recently was hardly understood is an area called the basal ganglia. It was thought to be connected to decision making. Different parts of the brain are able to control different body movements like winking (which medics call 'motor actions') or purely mental operations like noticing a drop of rain. In response to the things that happen around us ('stimuli'), these brain regions want to take control of our body or our attention. The basal ganglia decides which brain area gets control of the body or the attention, and in effect which action or mental operations actually happen. The basal ganglia is actually a bit like the operating system of a computer (like Linux or Windows XP). It similarly decides which application (computer program) gets to access the computers' output devices like the screen or the processor.
When the brain makes a decision, for example to look left rather than right, it does so only after processing lots of data. Suppose you are looking left and you hear something to your right. At some point the brain will decide to look right to see what the noise is. You don't do this the instant you hear the noise, but only after a certain amount of stimulus has been received by the brain, i.e. after some threshold has been reached. The basal ganglia was believed to perform this but it wasn't understood how ... and the Basal ganglia contained some very strange neurons not found in other parts of the brain.
The same type of decision problem is also studied in statistics. In statistics there is a particular procedure that's the best known method for making decisions on incoming data. It's described by a complicated mathematical formulae. Researchers, at Bristol and Sheffield, discovered that this formulae can be written in a way that is similar to a Neural Network: the computer scientist's version of that network of neurons like the water flowing through pipes. The formulae can be written as a series of simpler parts whose input and outputs are passed between each other. This is not very surprising, as all mathematical formulae can be written this way. What was surprising was that the resulting Neural Network looked remarkably similar to the wiring that was found in the basal ganglia. Could it be that the brain had, via the process of evolution, programmed itself to implement a complex statistical procedure?
Supporting the model
So this version of the statistical procedure acts as a model of the brain. Like a model of a plane we might build as a kid, it may be more or less an accurate copy of the real thing. Some things may be the same but others not be faithful.
Once scientists have a model of something then often then try to predict something which follows from the model and see whether this appears in the real world. This gives us confidence that the model is correct. We believe Newton's Laws of Motion as they allow us to predict where the planets are in the sky and to predict how objects travel on earth. Scientific models are a way to see into the future!
A particularly interesting part of the Neural Network model developed was that one of the groups of neurons in the model needed to perform a calculation called the exponential function. In other words if the input to the neuron was zero then the output would be one. If the input was 2 the output would be 4. If the input was 8, say, the output was 256. In maths terms if the input, x was greater than zero then the output would behave like 2^x. In other words it would grow very fast.
Remember that the basal ganglia had some strange neurons? Well the model predicts that the strange neurons should behave like the exponential function, and when this was checked they did! The biologists confirmed that they 'fire' (i.e. produce an output) even when no stimulus (i.e. input) was given to them. More to the point the output signals they produce grow very rapidly as the input signal increases and following the behaviour of the exponential function mentioned above.
So by understanding how to implement statistics using Neural Networks, the tools from Computer Science, we are led to a greater understanding of parts of the brain which were a mystery. The goal now is to use this model to predict what happens when this part of the brain is damaged. That way we might produce cures for the resulting diseases. In addition, there are still many aspects of this area of the brain which we do not understand. Scientists hope that similar techniques from Computer Science will help us increase our understanding of those too.
So we may not have created a computer that thinks yet, but the computers are helping us understand how we think!